Facial Expression Recognition Based On Deep Learning

With the rapid development of computer field, human feel so untouched that computer can have intelligence like human and have feelings like we do. The facial expression recognition is studied by more people as a key and hot research.In this thesis, the deep learning methods are used to recognize the facial expression. The facial expression method based on Principal Component Analysis(PCA) and Stacked Denoising Autoencoders(SDAE) and facial expression method based on Gabor wavelet and Deep Belief Networks(DBNs) are mainly studied.Firstly, a facial expression recognition method based on PCA and SDAE is proposed based on strong ability to learn the feature of SDAE, thus to apply the SDAE to the facial expression recognition. After the preprocessing, the dimension of faical images is reduced by PCA in a linear way. Then the feature is used for SDAE to learn the feature by a laywised way again. At the same time, the dimension of feature is reduced in a nonlinear way to get a lower dimension and better feature. Then the feature is used to classify. The comparative experiment results show that the proposed method can achieve higher facial recognition rate than some other deep learning and non-deep learning expression recognition methods in this thesis.Furthermore, because of the shortness to express the local character when the deep learning method learn the feature while the Gabor wavelet feature have a strong ability to extract the local feature, a novel facial expression method based on Gabor wavelet and DBNs is proposed in this paper. Firstly the Gabor wavelet transform is used to obtain the local feature of the facial expression, which is the Gabor wavelet feature of different scales and orientations based on the image pixel. And then it is used for the DBNs to learn the feature in a greed lay-wised way and then classify. The comparative experiment results show that the proposed method can learn the more effective features that combine the local and global character than some other deep learning and non-deep learning expression recognition methods in this thesis. Thus the proposed method can achieve higher facial expression recognition rate.At last, a facial expression recognition prototype system based on Deep Learning is designed and developed in this paper. There are two main lines that are facial expression recognition method based on PCA+SDAE and Gabor+DBNs in this system. And it includes data preprocessing, feature conduced and deep learning model modules. The testing results indicate that the system can recognize the facial expression effectively by the deep learning methods.